Knowledge compilation/pattern reasoning system

Abstract

A knowledge compilation/pattern reasoning system has two subsystems: a pattern generating subsystem for interpreting human representation of knowledge and physical models and transforming them into pattern data, and an executing subsystem for analyzing states by using the generated pattern data and decision criterion data. An operation system of the kind adopting pattern type reasoning may be loaded with only the executing subsystem which uses pattern data compiled by the pattern generating subsystem beforehand, for the purpose of implementing high-speed execution. Since the transformation of human representation of knowledge and physical models into pattern data is effected by the pattern generating subsystem which is independent of the executing system, human representation of knowledge and physical models and pattern data are desirably interfaced to each other.

What is claimed is: 
     
       1. A knowledge compilation/pattern reasoning system, comprising: a pattern generating subsystem for generating (a) state analysis pattern data including (a1) a comparision data sequence which describes a particular state based on a knowledge database, said knowledge database describing experiential knowledge and physical models necessary for analysis of a plurality of states, and   (a2) search tables for determining a next comparison pattern, and     (b) decision criterion data for state analysis composed of (b1) a first data sequence describing a decision characteristic associated with each of a plurality of input signals, and   (b2) a second data sequence describing an output characteristic of each input signal; and       an executing subsystem responsive to said plurality of input signals for analyzing said states by using the state analysis pattern data and the decision criterion data generated by said pattern generating subsystem, and said executing subsystem transmitting output signals matching said states,   wherein said pattern generating subsystem includes a patterning rule generating section for determining, using data loaded in said knowledge database, (1) a format of patterns to be generated and (2) states which individual patterns indicate, and said pattern rule generating section for storing said determined format and states in a patterning rule database;   a pattern data generating section for generating, by forward reasoning which uses the data loaded in said knowledge database, pattern data associated with the states which are indicated by the patterns stored in said patterning rule database, and said pattern data generating section for storing said pattern data in a pattern database; and   a pattern data verifying section for verifying, by backward reasoning which uses the data loaded in said knowledge database, individual pattern data stored in said pattern data base, and if any of said pattern data is not verified, said pattern data verifying section correcting said pattern data.     
     
     
       2. A system as claimed in claim 1, wherein said executing subsystem comprises means for changing, when an evaluation associated with state analysis is produced, the criterion data in matching relation to said evaluation, wherein said system adapts to a change in state. 
     
     
       3. A system as claimed in claim 1, wherein the first and second data sequences of the decision criterion data are a function of a characteristic function describing a characteristic of the input signals and a time delay having time as a factor. 
     
     
       4. A knowledge compilation/pattern reasoning system, comprising: a pattern generating subsystem for generating (a) state analysis pattern data including (a1) a comparison data sequence which describes a particular state based on a knowledge database, said knowledge database describing experiential knowledge and physical models necessary for analysis of a plurality of states, and   (a2) search tables for determining a next comparison pattern, and     (b) decision criterion data for state analysis composed of   (b1) a first data sequence describing a decision characteristic associated with each of a plurality of input signals, and   (b2) a second data sequence describing an output characteristic of each input signal; and     an executing subsystem responsive to said plurality of input signals for analyzing said states by using the state analysis pattern data and the decision criterion data generated by said pattern generating subsystem, and said executing subsystem transmitting output signals matching said states wherein said executing subsystem includes   an input normalizing section for normalizing individual input data based on input data normalizing tables and for outputting said normalized input data;   a pattern comparing section for sequentially calculating coincidence values by using (1) the normalized input data, (2) said comparison data sequence which is loaded in a pattern database generated by said pattern generating subsystem, said pattern data base is designated by said search tables which are described by pattern data, and (3) decision criterion data, and said pattern comparing section for outputting a pattern having a large coincidence value and said coincidence value;   an output combining section for calculating corrected coincidence values, individually associated with coincidence values from said pattern comparing section, by using a characteristic function, a time delay function or similar correcting function, for selecting a pattern having a large coincidence value and a pattern class for said entire system by using said corrected coincidence values, and for calculating participation values representative of extents to which individual decision criterion data have influenced selection of the pattern and the pattern class; and   a decision criterion changing section which is triggered upon input of evaluation data associated with output data of said output combining section, said decision criterion changing section for changing individual decision criterion data using (1) said evaluation data and (2) the participation values.   
     
     
       5. A system as claimed in claim 4, wherein said input normalizing section comprises input normalizers corresponding in number to said input signals and operable in parallel with each other; said pattern comparing section comprising pattern comparators corresponding in number to the number of decision criteria data and operable in parallel with each other.   
     
     
       6. A knowledge compilation/pattern reasoning system according to claim 4 wherein said pattern comparing section comprises: an input interface means for reading, during initialization, gravities of decision criterion data and, after initialization, sequentially reading input data to a coincidence value calculation means while sequentially outputting said gravities read during initialization to an integration means;   a pattern data interface means for reading, upon reception of a comparison pattern number from a pattern search means, search table data out of said pattern database for input to said pattern search means and, after initialization, sequentially reading said pattern data out of said pattern database for input to said coincidence value calculation means;   said coincidence value calculation means for generating a comparison function using the pattern data read via said pattern data interface means and an original function, for calculating coincidence values using the input data entered via said input interface means and said comparison function, and for outputting said coincidence values to said integration means;   said integration means having an integration buffer and counter buffer for clearing, upon reception of a comparison pattern number from said pattern search means, said integration buffer and said counter buffer and, after initialization, multiplying said coincidence values and said gravities read out of said coincidence value calculation means and said input interface means, respectively, for integrating results of multiplication and the gravities, for incrementing counter data, and, upon increment of the counter data to an upper limit, for transmitting a pattern number, for dividing an integrated value of the results of multiplication by an integrated value of the gravities, and for transmitting data representative of a result of division to said pattern search means; and   said pattern search means for selecting, upon reception of a pattern number, a pattern having a largest coincidence value and for storing said pattern in an internal buffer, for selecting a next comparison pattern based on said search tables stored in an internal buffer and for transmitting said next comparison pattern to said pattern data interface means, and, when said next comparison pattern is a particular pattern, for outputting a number of a pattern having a largest coincidence value.